diff options
Diffstat (limited to 'fs/aio.c')
| -rw-r--r-- | fs/aio.c | 1578 |
1 files changed, 550 insertions, 1028 deletions
@@ -8,6 +8,8 @@ * * See ../COPYING for licensing terms. */ +#define pr_fmt(fmt) "%s: " fmt, __func__ + #include <linux/kernel.h> #include <linux/init.h> #include <linux/errno.h> @@ -18,8 +20,6 @@ #include <linux/backing-dev.h> #include <linux/uio.h> -#define DEBUG 0 - #include <linux/sched.h> #include <linux/fs.h> #include <linux/file.h> @@ -39,11 +39,76 @@ #include <asm/kmap_types.h> #include <asm/uaccess.h> -#if DEBUG > 1 -#define dprintk printk -#else -#define dprintk(x...) do { ; } while (0) -#endif +#define AIO_RING_MAGIC 0xa10a10a1 +#define AIO_RING_COMPAT_FEATURES 1 +#define AIO_RING_INCOMPAT_FEATURES 0 +struct aio_ring { + unsigned id; /* kernel internal index number */ + unsigned nr; /* number of io_events */ + unsigned head; + unsigned tail; + + unsigned magic; + unsigned compat_features; + unsigned incompat_features; + unsigned header_length; /* size of aio_ring */ + + + struct io_event io_events[0]; +}; /* 128 bytes + ring size */ + +#define AIO_RING_PAGES 8 + +struct kioctx { + atomic_t users; + atomic_t dead; + + /* This needs improving */ + unsigned long user_id; + struct hlist_node list; + + /* + * This is what userspace passed to io_setup(), it's not used for + * anything but counting against the global max_reqs quota. + * + * The real limit is nr_events - 1, which will be larger (see + * aio_setup_ring()) + */ + unsigned max_reqs; + + /* Size of ringbuffer, in units of struct io_event */ + unsigned nr_events; + + unsigned long mmap_base; + unsigned long mmap_size; + + struct page **ring_pages; + long nr_pages; + + struct rcu_head rcu_head; + struct work_struct rcu_work; + + struct { + atomic_t reqs_active; + } ____cacheline_aligned_in_smp; + + struct { + spinlock_t ctx_lock; + struct list_head active_reqs; /* used for cancellation */ + } ____cacheline_aligned_in_smp; + + struct { + struct mutex ring_lock; + wait_queue_head_t wait; + } ____cacheline_aligned_in_smp; + + struct { + unsigned tail; + spinlock_t completion_lock; + } ____cacheline_aligned_in_smp; + + struct page *internal_pages[AIO_RING_PAGES]; +}; /*------ sysctl variables----*/ static DEFINE_SPINLOCK(aio_nr_lock); @@ -54,11 +119,6 @@ unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio request static struct kmem_cache *kiocb_cachep; static struct kmem_cache *kioctx_cachep; -static struct workqueue_struct *aio_wq; - -static void aio_kick_handler(struct work_struct *); -static void aio_queue_work(struct kioctx *); - /* aio_setup * Creates the slab caches used by the aio routines, panic on * failure as this is done early during the boot sequence. @@ -68,10 +128,7 @@ static int __init aio_setup(void) kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); - aio_wq = alloc_workqueue("aio", 0, 1); /* used to limit concurrency */ - BUG_ON(!aio_wq); - - pr_debug("aio_setup: sizeof(struct page) = %d\n", (int)sizeof(struct page)); + pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); return 0; } @@ -79,28 +136,23 @@ __initcall(aio_setup); static void aio_free_ring(struct kioctx *ctx) { - struct aio_ring_info *info = &ctx->ring_info; long i; - for (i=0; i<info->nr_pages; i++) - put_page(info->ring_pages[i]); + for (i = 0; i < ctx->nr_pages; i++) + put_page(ctx->ring_pages[i]); - if (info->mmap_size) { - BUG_ON(ctx->mm != current->mm); - vm_munmap(info->mmap_base, info->mmap_size); - } + if (ctx->mmap_size) + vm_munmap(ctx->mmap_base, ctx->mmap_size); - if (info->ring_pages && info->ring_pages != info->internal_pages) - kfree(info->ring_pages); - info->ring_pages = NULL; - info->nr = 0; + if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) + kfree(ctx->ring_pages); } static int aio_setup_ring(struct kioctx *ctx) { struct aio_ring *ring; - struct aio_ring_info *info = &ctx->ring_info; unsigned nr_events = ctx->max_reqs; + struct mm_struct *mm = current->mm; unsigned long size, populate; int nr_pages; @@ -116,46 +168,44 @@ static int aio_setup_ring(struct kioctx *ctx) nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event); - info->nr = 0; - info->ring_pages = info->internal_pages; + ctx->nr_events = 0; + ctx->ring_pages = ctx->internal_pages; if (nr_pages > AIO_RING_PAGES) { - info->ring_pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL); - if (!info->ring_pages) + ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *), + GFP_KERNEL); + if (!ctx->ring_pages) return -ENOMEM; } - info->mmap_size = nr_pages * PAGE_SIZE; - dprintk("attempting mmap of %lu bytes\n", info->mmap_size); - down_write(&ctx->mm->mmap_sem); - info->mmap_base = do_mmap_pgoff(NULL, 0, info->mmap_size, - PROT_READ|PROT_WRITE, - MAP_ANONYMOUS|MAP_PRIVATE, 0, - &populate); - if (IS_ERR((void *)info->mmap_base)) { - up_write(&ctx->mm->mmap_sem); - info->mmap_size = 0; + ctx->mmap_size = nr_pages * PAGE_SIZE; + pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size); + down_write(&mm->mmap_sem); + ctx->mmap_base = do_mmap_pgoff(NULL, 0, ctx->mmap_size, + PROT_READ|PROT_WRITE, + MAP_ANONYMOUS|MAP_PRIVATE, 0, &populate); + if (IS_ERR((void *)ctx->mmap_base)) { + up_write(&mm->mmap_sem); + ctx->mmap_size = 0; aio_free_ring(ctx); return -EAGAIN; } - dprintk("mmap address: 0x%08lx\n", info->mmap_base); - info->nr_pages = get_user_pages(current, ctx->mm, - info->mmap_base, nr_pages, - 1, 0, info->ring_pages, NULL); - up_write(&ctx->mm->mmap_sem); + pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); + ctx->nr_pages = get_user_pages(current, mm, ctx->mmap_base, nr_pages, + 1, 0, ctx->ring_pages, NULL); + up_write(&mm->mmap_sem); - if (unlikely(info->nr_pages != nr_pages)) { + if (unlikely(ctx->nr_pages != nr_pages)) { aio_free_ring(ctx); return -EAGAIN; } if (populate) - mm_populate(info->mmap_base, populate); + mm_populate(ctx->mmap_base, populate); - ctx->user_id = info->mmap_base; + ctx->user_id = ctx->mmap_base; + ctx->nr_events = nr_events; /* trusted copy */ - info->nr = nr_events; /* trusted copy */ - - ring = kmap_atomic(info->ring_pages[0]); + ring = kmap_atomic(ctx->ring_pages[0]); ring->nr = nr_events; /* user copy */ ring->id = ctx->user_id; ring->head = ring->tail = 0; @@ -164,72 +214,133 @@ static int aio_setup_ring(struct kioctx *ctx) ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; ring->header_length = sizeof(struct aio_ring); kunmap_atomic(ring); + flush_dcache_page(ctx->ring_pages[0]); return 0; } - -/* aio_ring_event: returns a pointer to the event at the given index from - * kmap_atomic(). Release the pointer with put_aio_ring_event(); - */ #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) -#define aio_ring_event(info, nr) ({ \ - unsigned pos = (nr) + AIO_EVENTS_OFFSET; \ - struct io_event *__event; \ - __event = kmap_atomic( \ - (info)->ring_pages[pos / AIO_EVENTS_PER_PAGE]); \ - __event += pos % AIO_EVENTS_PER_PAGE; \ - __event; \ -}) - -#define put_aio_ring_event(event) do { \ - struct io_event *__event = (event); \ - (void)__event; \ - kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK)); \ -} while(0) - -static void ctx_rcu_free(struct rcu_head *head) +void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel) +{ + struct kioctx *ctx = req->ki_ctx; + unsigned long flags; + + spin_lock_irqsave(&ctx->ctx_lock, flags); + + if (!req->ki_list.next) + list_add(&req->ki_list, &ctx->active_reqs); + + req->ki_cancel = cancel; + + spin_unlock_irqrestore(&ctx->ctx_lock, flags); +} +EXPORT_SYMBOL(kiocb_set_cancel_fn); + +static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb, + struct io_event *res) +{ + kiocb_cancel_fn *old, *cancel; + int ret = -EINVAL; + + /* + * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it + * actually has a cancel function, hence the cmpxchg() + */ + + cancel = ACCESS_ONCE(kiocb->ki_cancel); + do { + if (!cancel || cancel == KIOCB_CANCELLED) + return ret; + + old = cancel; + cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED); + } while (cancel != old); + + atomic_inc(&kiocb->ki_users); + spin_unlock_irq(&ctx->ctx_lock); + + memset(res, 0, sizeof(*res)); + res->obj = (u64)(unsigned long)kiocb->ki_obj.user; + res->data = kiocb->ki_user_data; + ret = cancel(kiocb, res); + + spin_lock_irq(&ctx->ctx_lock); + + return ret; +} + +static void free_ioctx_rcu(struct rcu_head *head) { struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); kmem_cache_free(kioctx_cachep, ctx); } -/* __put_ioctx - * Called when the last user of an aio context has gone away, - * and the struct needs to be freed. +/* + * When this function runs, the kioctx has been removed from the "hash table" + * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - + * now it's safe to cancel any that need to be. */ -static void __put_ioctx(struct kioctx *ctx) +static void free_ioctx(struct kioctx *ctx) { - unsigned nr_events = ctx->max_reqs; - BUG_ON(ctx->reqs_active); + struct aio_ring *ring; + struct io_event res; + struct kiocb *req; + unsigned head, avail; - cancel_delayed_work_sync(&ctx->wq); - aio_free_ring(ctx); - mmdrop(ctx->mm); - ctx->mm = NULL; - if (nr_events) { - spin_lock(&aio_nr_lock); - BUG_ON(aio_nr - nr_events > aio_nr); - aio_nr -= nr_events; - spin_unlock(&aio_nr_lock); + spin_lock_irq(&ctx->ctx_lock); + + while (!list_empty(&ctx->active_reqs)) { + req = list_first_entry(&ctx->active_reqs, + struct kiocb, ki_list); + + list_del_init(&req->ki_list); + kiocb_cancel(ctx, req, &res); } - pr_debug("__put_ioctx: freeing %p\n", ctx); - call_rcu(&ctx->rcu_head, ctx_rcu_free); -} -static inline int try_get_ioctx(struct kioctx *kioctx) -{ - return atomic_inc_not_zero(&kioctx->users); + spin_unlock_irq(&ctx->ctx_lock); + + ring = kmap_atomic(ctx->ring_pages[0]); + head = ring->head; + kunmap_atomic(ring); + + while (atomic_read(&ctx->reqs_active) > 0) { + wait_event(ctx->wait, head != ctx->tail); + + avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head; + + atomic_sub(avail, &ctx->reqs_active); + head += avail; + head %= ctx->nr_events; + } + + WARN_ON(atomic_read(&ctx->reqs_active) < 0); + + aio_free_ring(ctx); + + spin_lock(&aio_nr_lock); + BUG_ON(aio_nr - ctx->max_reqs > aio_nr); + aio_nr -= ctx->max_reqs; + spin_unlock(&aio_nr_lock); + + pr_debug("freeing %p\n", ctx); + + /* + * Here the call_rcu() is between the wait_event() for reqs_active to + * hit 0, and freeing the ioctx. + * + * aio_complete() decrements reqs_active, but it has to touch the ioctx + * after to issue a wakeup so we use rcu. + */ + call_rcu(&ctx->rcu_head, free_ioctx_rcu); } -static inline void put_ioctx(struct kioctx *kioctx) +static void put_ioctx(struct kioctx *ctx) { - BUG_ON(atomic_read(&kioctx->users) <= 0); - if (unlikely(atomic_dec_and_test(&kioctx->users))) - __put_ioctx(kioctx); + if (unlikely(atomic_dec_and_test(&ctx->users))) + free_ioctx(ctx); } /* ioctx_alloc @@ -237,7 +348,7 @@ static inline void put_ioctx(struct kioctx *kioctx) */ static struct kioctx *ioctx_alloc(unsigned nr_events) { - struct mm_struct *mm; + struct mm_struct *mm = current->mm; struct kioctx *ctx; int err = -ENOMEM; @@ -256,17 +367,15 @@ static struct kioctx *ioctx_alloc(unsigned nr_events) return ERR_PTR(-ENOMEM); ctx->max_reqs = nr_events; - mm = ctx->mm = current->mm; - atomic_inc(&mm->mm_count); atomic_set(&ctx->users, 2); + atomic_set(&ctx->dead, 0); spin_lock_init(&ctx->ctx_lock); - spin_lock_init(&ctx->ring_info.ring_lock); + spin_lock_init(&ctx->completion_lock); + mutex_init(&ctx->ring_lock); init_waitqueue_head(&ctx->wait); INIT_LIST_HEAD(&ctx->active_reqs); - INIT_LIST_HEAD(&ctx->run_list); - INIT_DELAYED_WORK(&ctx->wq, aio_kick_handler); if (aio_setup_ring(ctx) < 0) goto out_freectx; @@ -286,64 +395,56 @@ static struct kioctx *ioctx_alloc(unsigned nr_events) hlist_add_head_rcu(&ctx->list, &mm->ioctx_list); spin_unlock(&mm->ioctx_lock); - dprintk("aio: allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", - ctx, ctx->user_id, current->mm, ctx->ring_info.nr); + pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", + ctx, ctx->user_id, mm, ctx->nr_events); return ctx; out_cleanup: err = -EAGAIN; aio_free_ring(ctx); out_freectx: - mmdrop(mm); kmem_cache_free(kioctx_cachep, ctx); - dprintk("aio: error allocating ioctx %d\n", err); + pr_debug("error allocating ioctx %d\n", err); return ERR_PTR(err); } -/* kill_ctx - * Cancels all outstanding aio requests on an aio context. Used - * when the processes owning a context have all exited to encourage - * the rapid destruction of the kioctx. - */ -static void kill_ctx(struct kioctx *ctx) +static void kill_ioctx_work(struct work_struct *work) { - int (*cancel)(struct kiocb *, struct io_event *); - struct task_struct *tsk = current; - DECLARE_WAITQUEUE(wait, tsk); - struct io_event res; + struct kioctx *ctx = container_of(work, struct kioctx, rcu_work); - spin_lock_irq(&ctx->ctx_lock); - ctx->dead = 1; - while (!list_empty(&ctx->active_reqs)) { - struct list_head *pos = ctx->active_reqs.next; - struct kiocb *iocb = list_kiocb(pos); - list_del_init(&iocb->ki_list); - cancel = iocb->ki_cancel; - kiocbSetCancelled(iocb); - if (cancel) { - iocb->ki_users++; - spin_unlock_irq(&ctx->ctx_lock); - cancel(iocb, &res); - spin_lock_irq(&ctx->ctx_lock); - } - } + wake_up_all(&ctx->wait); + put_ioctx(ctx); +} - if (!ctx->reqs_active) - goto out; +static void kill_ioctx_rcu(struct rcu_head *head) +{ + struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); - add_wait_queue(&ctx->wait, &wait); - set_task_state(tsk, TASK_UNINTERRUPTIBLE); - while (ctx->reqs_active) { - spin_unlock_irq(&ctx->ctx_lock); - io_schedule(); - set_task_state(tsk, TASK_UNINTERRUPTIBLE); - spin_lock_irq(&ctx->ctx_lock); - } - __set_task_state(tsk, TASK_RUNNING); - remove_wait_queue(&ctx->wait, &wait); + INIT_WORK(&ctx->rcu_work, kill_ioctx_work); + schedule_work(&ctx->rcu_work); +} -out: - spin_unlock_irq(&ctx->ctx_lock); +/* kill_ioctx + * Cancels all outstanding aio requests on an aio context. Used + * when the processes owning a context have all exited to encourage + * the rapid destruction of the kioctx. + */ +static void kill_ioctx(struct kioctx *ctx) +{ + if (!atomic_xchg(&ctx->dead, 1)) { + hlist_del_rcu(&ctx->list); + /* Between hlist_del_rcu() and dropping the initial ref */ + synchronize_rcu(); + + /* + * We can't punt to workqueue here because put_ioctx() -> + * free_ioctx() will unmap the ringbuffer, and that has to be + * done in the original process's context. kill_ioctx_rcu/work() + * exist for exit_aio(), as in that path free_ioctx() won't do + * the unmap. + */ + kill_ioctx_work(&ctx->rcu_work); + } } /* wait_on_sync_kiocb: @@ -351,9 +452,9 @@ out: */ ssize_t wait_on_sync_kiocb(struct kiocb *iocb) { - while (iocb->ki_users) { + while (atomic_read(&iocb->ki_users)) { set_current_state(TASK_UNINTERRUPTIBLE); - if (!iocb->ki_users) + if (!atomic_read(&iocb->ki_users)) break; io_schedule(); } @@ -362,28 +463,26 @@ ssize_t wait_on_sync_kiocb(struct kiocb *iocb) } EXPORT_SYMBOL(wait_on_sync_kiocb); -/* exit_aio: called when the last user of mm goes away. At this point, - * there is no way for any new requests to be submited or any of the - * io_* syscalls to be called on the context. However, there may be - * outstanding requests which hold references to the context; as they - * go away, they will call put_ioctx and release any pinned memory - * associated with the request (held via struct page * references). +/* + * exit_aio: called when the last user of mm goes away. At this point, there is + * no way for any new requests to be submited or any of the io_* syscalls to be + * called on the context. + * + * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on + * them. */ void exit_aio(struct mm_struct *mm) { struct kioctx *ctx; + struct hlist_node *n; - while (!hlist_empty(&mm->ioctx_list)) { - ctx = hlist_entry(mm->ioctx_list.first, struct kioctx, list); - hlist_del_rcu(&ctx->list); - - kill_ctx(ctx); - + hlist_for_each_entry_safe(ctx, n, &mm->ioctx_list, list) { if (1 != atomic_read(&ctx->users)) printk(KERN_DEBUG "exit_aio:ioctx still alive: %d %d %d\n", - atomic_read(&ctx->users), ctx->dead, - ctx->reqs_active); + atomic_read(&ctx->users), + atomic_read(&ctx->dead), + atomic_read(&ctx->reqs_active)); /* * We don't need to bother with munmap() here - * exit_mmap(mm) is coming and it'll unmap everything. @@ -391,150 +490,53 @@ void exit_aio(struct mm_struct *mm) * as indicator that it needs to unmap the area, * just set it to 0; aio_free_ring() is the only * place that uses ->mmap_size, so it's safe. - * That way we get all munmap done to current->mm - - * all other callers have ctx->mm == current->mm. */ - ctx->ring_info.mmap_size = 0; - put_ioctx(ctx); + ctx->mmap_size = 0; + + if (!atomic_xchg(&ctx->dead, 1)) { + hlist_del_rcu(&ctx->list); + call_rcu(&ctx->rcu_head, kill_ioctx_rcu); + } } } /* aio_get_req - * Allocate a slot for an aio request. Increments the users count + * Allocate a slot for an aio request. Increments the ki_users count * of the kioctx so that the kioctx stays around until all requests are * complete. Returns NULL if no requests are free. * - * Returns with kiocb->users set to 2. The io submit code path holds + * Returns with kiocb->ki_users set to 2. The io submit code path holds * an extra reference while submitting the i/o. * This prevents races between the aio code path referencing the * req (after submitting it) and aio_complete() freeing the req. */ -static struct kiocb *__aio_get_req(struct kioctx *ctx) +static inline struct kiocb *aio_get_req(struct kioctx *ctx) { - struct kiocb *req = NULL; + struct kiocb *req; - req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL); - if (unlikely(!req)) + if (atomic_read(&ctx->reqs_active) >= ctx->nr_events) return NULL; - req->ki_flags = 0; - req->ki_users = 2; - req->ki_key = 0; - req->ki_ctx = ctx; - req->ki_cancel = NULL; - req->ki_retry = NULL; - req->ki_dtor = NULL; - req->private = NULL; - req->ki_iovec = NULL; - INIT_LIST_HEAD(&req->ki_run_list); - req->ki_eventfd = NULL; - - return req; -} - -/* - * struct kiocb's are allocated in batches to reduce the number of - * times the ctx lock is acquired and released. - */ -#define KIOCB_BATCH_SIZE 32L -struct kiocb_batch { - struct list_head head; - long count; /* number of requests left to allocate */ -}; - -static void kiocb_batch_init(struct kiocb_batch *batch, long total) -{ - INIT_LIST_HEAD(&batch->head); - batch->count = total; -} - -static void kiocb_batch_free(struct kioctx *ctx, struct kiocb_batch *batch) -{ - struct kiocb *req, *n; - - if (list_empty(&batch->head)) - return; - - spin_lock_irq(&ctx->ctx_lock); - list_for_each_entry_safe(req, n, &batch->head, ki_batch) { - list_del(&req->ki_batch); - list_del(&req->ki_list); - kmem_cache_free(kiocb_cachep, req); - ctx->reqs_active--; - } - if (unlikely(!ctx->reqs_active && ctx->dead)) - wake_up_all(&ctx->wait); - spin_unlock_irq(&ctx->ctx_lock); -} - -/* - * Allocate a batch of kiocbs. This avoids taking and dropping the - * context lock a lot during setup. - */ -static int kiocb_batch_refill(struct kioctx *ctx, struct kiocb_batch *batch) -{ - unsigned short allocated, to_alloc; - long avail; - struct kiocb *req, *n; - struct aio_ring *ring; - - to_alloc = min(batch->count, KIOCB_BATCH_SIZE); - for (allocated = 0; allocated < to_alloc; allocated++) { - req = __aio_get_req(ctx); - if (!req) - /* allocation failed, go with what we've got */ - break; - list_add(&req->ki_batch, &batch->head); - } - - if (allocated == 0) - goto out; - - spin_lock_irq(&ctx->ctx_lock); - ring = kmap_atomic(ctx->ring_info.ring_pages[0]); - - avail = aio_ring_avail(&ctx->ring_info, ring) - ctx->reqs_active; - BUG_ON(avail < 0); - if (avail < allocated) { - /* Trim back the number of requests. */ - list_for_each_entry_safe(req, n, &batch->head, ki_batch) { - list_del(&req->ki_batch); - kmem_cache_free(kiocb_cachep, req); - if (--allocated <= avail) - break; - } - } - - batch->count -= allocated; - list_for_each_entry(req, &batch->head, ki_batch) { - list_add(&req->ki_list, &ctx->active_reqs); - ctx->reqs_active++; - } + if (atomic_inc_return(&ctx->reqs_active) > ctx->nr_events - 1) + goto out_put; - kunmap_atomic(ring); - spin_unlock_irq(&ctx->ctx_lock); - -out: - return allocated; -} + req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO); + if (unlikely(!req)) + goto out_put; -static inline struct kiocb *aio_get_req(struct kioctx *ctx, - struct kiocb_batch *batch) -{ - struct kiocb *req; + atomic_set(&req->ki_users, 2); + req->ki_ctx = ctx; - if (list_empty(&batch->head)) - if (kiocb_batch_refill(ctx, batch) == 0) - return NULL; - req = list_first_entry(&batch->head, struct kiocb, ki_batch); - list_del(&req->ki_batch); return req; +out_put: + atomic_dec(&ctx->reqs_active); + return NULL; } -static inline void really_put_req(struct kioctx *ctx, struct kiocb *req) +static void kiocb_free(struct kiocb *req) { - assert_spin_locked(&ctx->ctx_lock); - + if (req->ki_filp) + fput(req->ki_filp); if (req->ki_eventfd != NULL) eventfd_ctx_put(req->ki_eventfd); if (req->ki_dtor) @@ -542,48 +544,12 @@ static inline void really_put_req(struct kioctx *ctx, struct kiocb *req) if (req->ki_iovec != &req->ki_inline_vec) kfree(req->ki_iovec); kmem_cache_free(kiocb_cachep, req); - ctx->reqs_active--; - - if (unlikely(!ctx->reqs_active && ctx->dead)) - wake_up_all(&ctx->wait); } -/* __aio_put_req - * Returns true if this put was the last user of the request. - */ -static int __aio_put_req(struct kioctx *ctx, struct kiocb *req) +void aio_put_req(struct kiocb *req) { - dprintk(KERN_DEBUG "aio_put(%p): f_count=%ld\n", - req, atomic_long_read(&req->ki_filp->f_count)); - - assert_spin_locked(&ctx->ctx_lock); - - req->ki_users--; - BUG_ON(req->ki_users < 0); - if (likely(req->ki_users)) - return 0; - list_del(&req->ki_list); /* remove from active_reqs */ - req->ki_cancel = NULL; - req->ki_retry = NULL; - - fput(req->ki_filp); - req->ki_filp = NULL; - really_put_req(ctx, req); - return 1; -} - -/* aio_put_req - * Returns true if this put was the last user of the kiocb, - * false if the request is still in use. - */ -int aio_put_req(struct kiocb *req) -{ - struct kioctx *ctx = req->ki_ctx; - int ret; - spin_lock_irq(&ctx->ctx_lock); - ret = __aio_put_req(ctx, req); - spin_unlock_irq(&ctx->ctx_lock); - return ret; + if (atomic_dec_and_test(&req->ki_users)) + kiocb_free(req); } EXPORT_SYMBOL(aio_put_req); @@ -595,13 +561,8 @@ static struct kioctx *lookup_ioctx(unsigned long ctx_id) rcu_read_lock(); hlist_for_each_entry_rcu(ctx, &mm->ioctx_list, list) { - /* - * RCU protects us against accessing freed memory but - * we have to be careful not to get a reference when the - * reference count already dropped to 0 (ctx->dead test - * is unreliable because of races). - */ - if (ctx->user_id == ctx_id && !ctx->dead && try_get_ioctx(ctx)){ + if (ctx->user_id == ctx_id) { + atomic_inc(&ctx->users); ret = ctx; break; } @@ -611,295 +572,16 @@ static struct kioctx *lookup_ioctx(unsigned long ctx_id) return ret; } -/* - * Queue up a kiocb to be retried. Assumes that the kiocb - * has already been marked as kicked, and places it on - * the retry run list for the corresponding ioctx, if it - * isn't already queued. Returns 1 if it actually queued - * the kiocb (to tell the caller to activate the work - * queue to process it), or 0, if it found that it was - * already queued. - */ -static inline int __queue_kicked_iocb(struct kiocb *iocb) -{ - struct kioctx *ctx = iocb->ki_ctx; - - assert_spin_locked(&ctx->ctx_lock); - - if (list_empty(&iocb->ki_run_list)) { - list_add_tail(&iocb->ki_run_list, - &ctx->run_list); - return 1; - } - return 0; -} - -/* aio_run_iocb - * This is the core aio execution routine. It is - * invoked both for initial i/o submission and - * subsequent retries via the aio_kick_handler. - * Expects to be invoked with iocb->ki_ctx->lock - * already held. The lock is released and reacquired - * as needed during processing. - * - * Calls the iocb retry method (already setup for the - * iocb on initial submission) for operation specific - * handling, but takes care of most of common retry - * execution details for a given iocb. The retry method - * needs to be non-blocking as far as possible, to avoid - * holding up other iocbs waiting to be serviced by the - * retry kernel thread. - * - * The trickier parts in this code have to do with - * ensuring that only one retry instance is in progress - * for a given iocb at any time. Providing that guarantee - * simplifies the coding of individual aio operations as - * it avoids various potential races. - */ -static ssize_t aio_run_iocb(struct kiocb *iocb) -{ - struct kioctx *ctx = iocb->ki_ctx; - ssize_t (*retry)(struct kiocb *); - ssize_t ret; - - if (!(retry = iocb->ki_retry)) { - printk("aio_run_iocb: iocb->ki_retry = NULL\n"); - return 0; - } - - /* - * We don't want the next retry iteration for this - * operation to start until this one has returned and - * updated the iocb state. However, wait_queue functions - * can trigger a kick_iocb from interrupt context in the - * meantime, indicating that data is available for the next - * iteration. We want to remember that and enable the - * next retry iteration _after_ we are through with - * this one. - * - * So, in order to be able to register a "kick", but - * prevent it from being queued now, we clear the kick - * flag, but make the kick code *think* that the iocb is - * still on the run list until we are actually done. - * When we are done with this iteration, we check if - * the iocb was kicked in the meantime and if so, queue - * it up afresh. - */ - - kiocbClearKicked(iocb); - - /* - * This is so that aio_complete knows it doesn't need to - * pull the iocb off the run list (We can't just call - * INIT_LIST_HEAD because we don't want a kick_iocb to - * queue this on the run list yet) - */ - iocb->ki_run_list.next = iocb->ki_run_list.prev = NULL; - spin_unlock_irq(&ctx->ctx_lock); - - /* Quit retrying if the i/o has been cancelled */ - if (kiocbIsCancelled(iocb)) { - ret = -EINTR; - aio_complete(iocb, ret, 0); - /* must not access the iocb after this */ - goto out; - } - - /* - * Now we are all set to call the retry method in async - * context. - */ - ret = retry(iocb); - - if (ret != -EIOCBRETRY && ret != -EIOCBQUEUED) { - /* - * There's no easy way to restart the syscall since other AIO's - * may be already running. Just fail this IO with EINTR. - */ - if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || - ret == -ERESTARTNOHAND || ret == -ERESTART_RESTARTBLOCK)) - ret = -EINTR; - aio_complete(iocb, ret, 0); - } -out: - spin_lock_irq(&ctx->ctx_lock); - - if (-EIOCBRETRY == ret) { - /* - * OK, now that we are done with this iteration - * and know that there is more left to go, - * this is where we let go so that a subsequent - * "kick" can start the next iteration - */ - - /* will make __queue_kicked_iocb succeed from here on */ - INIT_LIST_HEAD(&iocb->ki_run_list); - /* we must queue the next iteration ourselves, if it - * has already been kicked */ - if (kiocbIsKicked(iocb)) { - __queue_kicked_iocb(iocb); - - /* - * __queue_kicked_iocb will always return 1 here, because - * iocb->ki_run_list is empty at this point so it should - * be safe to unconditionally queue the context into the - * work queue. - */ - aio_queue_work(ctx); - } - } - return ret; -} - -/* - * __aio_run_iocbs: - * Process all pending retries queued on the ioctx - * run list. - * Assumes it is operating within the aio issuer's mm - * context. - */ -static int __aio_run_iocbs(struct kioctx *ctx) -{ - struct kiocb *iocb; - struct list_head run_list; - - assert_spin_locked(&ctx->ctx_lock); - - list_replace_init(&ctx->run_list, &run_list); - while (!list_empty(&run_list)) { - iocb = list_entry(run_list.next, struct kiocb, - ki_run_list); - list_del(&iocb->ki_run_list); - /* - * Hold an extra reference while retrying i/o. - */ - iocb->ki_users++; /* grab extra reference */ - aio_run_iocb(iocb); - __aio_put_req(ctx, iocb); - } - if (!list_empty(&ctx->run_list)) - return 1; - return 0; -} - -static void aio_queue_work(struct kioctx * ctx) -{ - unsigned long timeout; - /* - * if someone is waiting, get the work started right - * away, otherwise, use a longer delay - */ - smp_mb(); - if (waitqueue_active(&ctx->wait)) - timeout = 1; - else - timeout = HZ/10; - queue_delayed_work(aio_wq, &ctx->wq, timeout); -} - -/* - * aio_run_all_iocbs: - * Process all pending retries queued on the ioctx - * run list, and keep running them until the list - * stays empty. - * Assumes it is operating within the aio issuer's mm context. - */ -static inline void aio_run_all_iocbs(struct kioctx *ctx) -{ - spin_lock_irq(&ctx->ctx_lock); - while (__aio_run_iocbs(ctx)) - ; - spin_unlock_irq(&ctx->ctx_lock); -} - -/* - * aio_kick_handler: - * Work queue handler triggered to process pending - * retries on an ioctx. Takes on the aio issuer's - * mm context before running the iocbs, so that - * copy_xxx_user operates on the issuer's address - * space. - * Run on aiod's context. - */ -static void aio_kick_handler(struct work_struct *work) -{ - struct kioctx *ctx = container_of(work, struct kioctx, wq.work); - mm_segment_t oldfs = get_fs(); - struct mm_struct *mm; - int requeue; - - set_fs(USER_DS); - use_mm(ctx->mm); - spin_lock_irq(&ctx->ctx_lock); - requeue =__aio_run_iocbs(ctx); - mm = ctx->mm; - spin_unlock_irq(&ctx->ctx_lock); - unuse_mm(mm); - set_fs(oldfs); - /* - * we're in a worker thread already; no point using non-zero delay - */ - if (requeue) - queue_delayed_work(aio_wq, &ctx->wq, 0); -} - - -/* - * Called by kick_iocb to queue the kiocb for retry - * and if required activate the aio work queue to process - * it - */ -static void try_queue_kicked_iocb(struct kiocb *iocb) -{ - struct kioctx *ctx = iocb->ki_ctx; - unsigned long flags; - int run = 0; - - spin_lock_irqsave(&ctx->ctx_lock, flags); - /* set this inside the lock so that we can't race with aio_run_iocb() - * testing it and putting the iocb on the run list under the lock */ - if (!kiocbTryKick(iocb)) - run = __queue_kicked_iocb(iocb); - spin_unlock_irqrestore(&ctx->ctx_lock, flags); - if (run) - aio_queue_work(ctx); -} - -/* - * kick_iocb: - * Called typically from a wait queue callback context - * to trigger a retry of the iocb. - * The retry is usually executed by aio workqueue - * threads (See aio_kick_handler). - */ -void kick_iocb(struct kiocb *iocb) -{ - /* sync iocbs are easy: they can only ever be executing from a - * single context. */ - if (is_sync_kiocb(iocb)) { - kiocbSetKicked(iocb); - wake_up_process(iocb->ki_obj.tsk); - return; - } - - try_queue_kicked_iocb(iocb); -} -EXPORT_SYMBOL(kick_iocb); - /* aio_complete * Called when the io request on the given iocb is complete. - * Returns true if this is the last user of the request. The - * only other user of the request can be the cancellation code. */ -int aio_complete(struct kiocb *iocb, long res, long res2) +void aio_complete(struct kiocb *iocb, long res, long res2) { struct kioctx *ctx = iocb->ki_ctx; - struct aio_ring_info *info; struct aio_ring *ring; - struct io_event *event; + struct io_event *ev_page, *event; unsigned long flags; - unsigned long tail; - int ret; + unsigned tail, pos; /* * Special case handling for sync iocbs: @@ -909,61 +591,81 @@ int aio_complete(struct kiocb *iocb, long res, long res2) * - the sync task helpfully left a reference to itself in the iocb */ if (is_sync_kiocb(iocb)) { - BUG_ON(iocb->ki_users != 1); + BUG_ON(atomic_read(&iocb->ki_users) != 1); iocb->ki_user_data = res; - iocb->ki_users = 0; + atomic_set(&iocb->ki_users, 0); wake_up_process(iocb->ki_obj.tsk); - return 1; + return; } - info = &ctx->ring_info; - - /* add a completion event to the ring buffer. - * must be done holding ctx->ctx_lock to prevent - * other code from messing with the tail - * pointer since we might be called from irq - * context. + /* + * Take rcu_read_lock() in case the kioctx is being destroyed, as we + * need to issue a wakeup after decrementing req |